Abstract

Active learning is an important field of machine learning and it is becoming more widely used in case of problems where labeling the examples in the training data set is expensive. In this paper we present a clustering-based algorithm used in the Active Learning Challenge (http://www.causality.inf.ethz.ch/activelearning.php). The algorithm is based on graph clustering with normalized cuts, and uses k-means to extract representative points from the data and approximate spectral clustering for efficiently performing the computations.